Considerations in the Production Process of High-Frequency PCBs: Station by Station

Every stage of a high-frequency build has a window, and the window is the consideration. “Control the etching process carefully” is not a specification; “line width ±10 % with ±25 µm capability, measured on a coupon at the panel centre and edge” is. This article walks the twelve stations in order — cut, inner layer, lamination, drilling, desmear and electroless copper, pattern plating, solder mask, finish, profiling, test, final inspection — and gives each one the four numbers that define it: temperature, pressure, tolerance and time, together with the failure that appears when the window is missed.

Considerations in the Production Process of High-Frequency PCBs

How to Read a Process Window

A process window has four dimensions, and a station is only controlled when all four are stated. Temperature without time means nothing, because a press cycle that reaches 200 °C for two minutes and one that holds it for ninety minutes are different processes. Pressure without ramp rate means nothing, because the flow of a bonding film depends on how fast the stack is compressed. Tolerance without a measurement location means nothing, because the centre and the edge of a 600 mm panel do not etch at the same rate.

The twelve stations below are the standard route for a multilayer high-frequency board. Each is presented with the window that matters and the failure that follows when the window is missed; the order is the order of the shop floor, because a fault introduced at station four is usually detected at station nine, which is why the detection column is as important as the window column.

StationGoverning windowTypical valueWhat fails if it is missed
1. Engineering / MIStack and impedance model agreementCoupon per layer, per targetA board that is 20 % out and unfixable
2. CuttingDimensional release before imaging3–5 mm rough cut, relax before finish cutScale error that propagates through every layer
3. Inner-layer image and etchLine width, and its uniformity across the panel±10 % or ±25 µm, whichever is tighterImpedance scatter, opens on fine lines
4. LaminationPeak temperature, pressure, ramp, vacuum170–300 °C depending on system; 200–400 psiThickness loss, slip, delamination
5. DrillingHole position, diameter, burr±25 µm position, ±50 µm diameterBarrel breakout, intermittent opens
6. Desmear and electroless copperChemistry per laminate, seed thickness0.3–0.5 µm seed, 45–50 °CVoided barrel, field failure after cycling
7. Pattern platingCurrent density and throwing power1.5–3 A/dm², surface 40–50 µmThin barrel, over-plated fine lines
8. Solder maskThickness over impedance traces, cure20–30 µm, 150 °C / 60 minImpedance 1–3 Ω low, poor adhesion after reflow
9. Surface finishMetal system and thicknessOSP 0.2–0.5 µm; ENIG Ni 3–6 µmLaunch loss, solderability, shelf life
10. ProfilingRouting method and edge tolerance±0.1 mm, no V-score on PTFEDelamination at the score line
11. Electrical testContinuity and isolation, four-wire100 % on high-value buildsEscapes that fail at assembly
12. Final inspectionCoupon data, cross-section, cleanlinessIPC-A-600G, IPC-6015No evidence trail when a lot is questioned

Station 1 — Engineering: The Window That Cannot Be Changed Later

Before anything is cut, three documents have to agree: the stack drawing, the impedance model, and the coupon geometry. The failure mode when they do not is specific and common — the model omits a bonding layer, and every board from that release is 20 % high in impedance with no process fault to find.

The window here is documentation, not physics. What has to be in it: dielectric thickness per layer with its tolerance, the copper foil class per face, the impedance target and the exact coupon geometry per layer, and the acceptance criteria. If a 0.10 mm bonding film between two cores is not in the model, a 50 Ω design becomes 61 Ω; if the foil class is not stated, the same drawing produces 0.19 or 0.145 dB/in of copper loss depending on what purchasing buys. The design-side version of this station is set out in high-frequency PCB design steps.

Station 2 — Cutting: Where Scale Is Set

High-frequency cores arrive as 600×1200 mm sheets, and the panel is cut from them. Two details matter more than the saw. The first is that PTFE expands measurably with temperature: a 600 mm TLY-5 panel changes length by about 0.31 mm over a 20 °C excursion, so a panel cut and immediately imaged at a different temperature than it will be pressed at carries a scale error before any process error is added.

The second is that the cut has to be a two-stage operation. A rough cut 3–5 mm oversize lets the sheet relieve its internal stress, and the finish cut to final dimension follows. Cutting to final size in one pass leaves the panel to move afterwards, and the movement appears as a registration error at lamination that nobody can attribute. Neither laser cutting nor water-jet is appropriate for PTFE: the first decomposes the material above about 400 °C and releases hydrogen fluoride, the second leaves an edge that will not plate reliably.

Station 3 — Inner-Layer Imaging and Etching

Inner-layer imaging on a high-frequency board is a tolerance operation, not a resolution operation. The number that matters is not the finest line the etcher can make but the variation of a 0.5 mm line across the panel, because that variation is impedance.

ParameterWindowEffect on the finished board
LDI pattern placement±5 µmLayer-to-layer registration budget, not line width
Line width tolerance±10 % or ±25 µm±2.8 % impedance at 0.59 mm; ±5.5 % at 0.254 mm
Etch factor1.3–1.5 for 1 ozUndercut of 23 µm per side at EF 1.5 on 1 oz
Panel uniformity≤10 % centre to edgeDense arrays etch narrow, panel edges etch wide
Etch compensation≥2× copper thicknessApplied in CAM to the drawn image, per layer

The uniformity row is the one that is usually unmanaged. Etch rate depends on fresh chemistry reaching the surface, so a dense array depletes the etchant locally and etches narrower, while an open border etches wider. The consequence for a high-frequency board is that the same drawn width finishes 10 % apart across one panel, which is an impedance spread no coupon in the border will reveal. The controls are placement — coupons in a copper-balanced area, not in the border — and measurement at both centre and edge. The chemistry side of the same step is covered in inner-layer circuit fabrication.

Station 4 — Lamination: Three Systems, Three Cycles

Lamination is where a high-frequency build differs most from a standard multilayer, and where a single mistake is most expensive, because everything downstream is built on the pressed stack. The windows are set by the bonding system rather than by the cores.

SystemPeak temperatureRampPressure / vacuumHold
FR-4170–190 °C1.5–2.5 °C/min200–350 psi, <50 mbar60–90 min above 170 °C
RO4000 series190–200 °C2–3 °C/min200–400 psi, <50 mbar60–120 min above 190 °C
PTFE with FEP film290–315 °C1.5–2 °C/min, slow cool200–500 psi, <30 mbar30–60 min above 290 °C

Three consequences have to be in the process plan. The cycles cannot be combined: a stack that contains both FR-4 and PTFE needs two press operations, and the second one re-heats everything already cured. Prepreg shrinks 0.02–0.05 % on cure, which is 0.12–0.30 mm across a 600 mm panel and has to be compensated per layer rather than absorbed. And PTFE cold-flows: 4 % compression moves a 50 Ω line to 48.8 Ω, so the pressure window is also an impedance window.

Thickness after press is the acceptance number: ±10 % of nominal per dielectric layer, measured on a cross-section, not on the stack height. A stack that is right in total but wrong in distribution still fails, because the impedance of each layer depends on its own dielectric. The full treatment of flow, slip and bonding films is in lamination in high-frequency PCB manufacturing.

Station 5 — Drilling: Position, Diameter, Burr

Drilling sets three independent numbers: where the hole is, how big it is, and what the exit looks like. On a high-frequency board all three are harder than on FR-4, and the reason is that the laminate is soft and does not conduct heat away from the tool edge.

ParameterFR-4PTFENote
Position tolerance±25 µm±25 µm, target ±15 µmAgainst the inner-layer target, via X-ray
Finished diameter±50 µm±50 µmFinished hole = drill − 2 × 25 µm of copper
Aspect ratio ceiling8:1 to 10:1about 6:1Non-wetting wall plus 237 ppm/°C Z expansion
Bit life2,000–3,000 hits500–1,000 hitsBurr grows before diameter does
Burr at exit≤25 µm≤25 µm, backing board mandatoryA burr is a short waiting for assembly

The aspect ratio ceiling deserves emphasis because it converts directly into a design constraint. At 6:1 on a 1.6 mm board, the smallest reliable finished hole is about 0.27 mm; ask for 0.2 mm and the plating has to throw into a hole whose wall will not wet, and the failure will be a barrel that passes continuity and opens after thermal cycling. Parameter detail, including the effect of retract and hit count on burr height, is in drilling high-frequency PCB.

Station 6 — Desmear and Electroless Copper: The Invisible Step

Desmear removes the resin smear left by drilling so that the electroless copper can contact the inner-layer pad. On epoxy laminates this is a permanganate process. On PTFE it is not: permanganate does not attack the smear, so the only routes are sodium naphthalenide or plasma.

The window: sodium naphthalenide treatment of a few minutes followed by a thorough rinse and neutralisation, or an oxygen plasma cycle. Either way the result has to be verified by cross-section, because the failure mode is invisible to electrical test. Electroless copper then deposits 0.3–0.5 µm of seed at 45–50 °C at roughly 2–3 µm/h; too thin a seed and the subsequent pattern plating cannot bridge the hole, too thick and the fine-line resist lifts.

The failure that matters is a barrel with a void in it. A 10 % void passes continuity, passes a thermal shock if it is on one side only, and opens in the field after a few hundred cycles. This is why the cross-section is the only credible inspection for this station and why electroless copper plating is treated as a critical process rather than a preparation step.

Station 7 — Pattern Plating: Thickness Where You Cannot See It

Pattern plating builds the barrel and the surface trace at the same time, which means the two requirements compete. The barrel needs enough copper to survive thermal expansion; the surface needs not so much copper that a 0.15 mm line closes up.

ParameterWindowReason
Current density1.5–3 A/dm²Above this the surface over-plates before the barrel is filled
Surface copper after plating40–50 µm total35 µm base plus 20–25 µm of plating at 5:1
Barrel thickness≥20 µm, ≥25 µm for PTFEZ expansion of 82 µm at reflow on 1.6 mm 5880
Plating ductility≥20 % elongationThe barrel has to stretch with the laminate
Line width after plating+10–25 µm on outer layersThe same image is 0.1 mm inside and 0.13 mm outside

The last row is why impedance coupons are layer-specific: an outer-layer coupon that is copied from an inner-layer design measures a different line. It is also why the outer-layer etch compensation is different from the inner-layer value, and why a single “±25 µm” on the drawing is not enough information to build the board. The chemistry and the panel-level current distribution are covered in plating high-frequency PCB.

Station 8 — Solder Mask: 25 µm of Dielectric on the Line

Solder mask is applied 20–30 µm thick over a trace that sits in the field of a microstrip. Its Dk is 3.5–4.5 depending on colour, and white legend ink is worse at about 6.5. The electrical result is an impedance drop of 1–3 Ω and a small increase in loss, both of which are absent from a coupon measured with the mask opened.

The window is therefore split. Cure is a standard 150 °C for about 60 minutes, or UV plus thermal for a photoimageable mask. Thickness is where the design leverage sits: keep it at the low end on impedance-critical layers, define mask openings over matched lines and coplanar gaps, and keep legend off any transmission line. A mask opening costs nothing in the artwork and recovers most of the impedance error. The application detail — mesh, viscosity, cure profile — is in solder mask application.

Station 9 — Surface Finish: The Last 0.1 dB

Finish selection on a high-frequency board is decided by two numbers: the skin depth at the operating frequency and the thickness of whatever metal the finish adds. At 10 GHz the skin depth is 0.65 µm in copper and 1.33 µm in nickel, so an ENIG layer of 3–6 µm of nickel puts the current in a metal with about four times the resistivity of copper.

The window per finish: OSP 0.2–0.5 µm of organic film with no metal added, immersion silver 0.1–0.3 µm, ENIG with 3–6 µm nickel and 0.05–0.1 µm gold, ENEPIG adding 0.05–0.1 µm of palladium. For microwave work the default is OSP or immersion silver, with ENIG reserved for fine pitch and wire bonding. The station-by-station chemistry and the IPC thickness windows are in surface treatment in high-frequency PCB production.

Station 10 — Profiling: Where Soft Laminates Are Lost

Routing is preferred over V-scoring on any PTFE build. A V-score cuts a stress concentration into a laminate that is soft and has a Z-axis CTE of 237 ppm/°C, and the delamination that follows appears at reflow and is usually blamed on lamination. Routing to ±0.1 mm with a 2 mm minimum web between boards avoids the issue.

The other number at this station is edge quality. A routed PTFE edge has a slight fuzz that does not affect function but does affect the solder mask adhesion at the edge and the appearance of a cavity drop-in. Deburring by hand is common and is also where a 0.5 mm thin panel gets bent; a dedicated edge-finishing step is cheaper than the scrap it prevents.

Station 11 — Electrical Test: What It Cannot See

Continuity and isolation testing verifies that the netlist is realised. It does not verify impedance, does not verify the barrel wall, and does not verify loss. On a high-frequency build that gap has to be closed with coupons and cross-sections, which is why a test strategy that consists only of a flying-probe pass is not evidence of a good board.

The practical arrangement: 100 % continuity and isolation on the finished panel with a four-wire measurement on power nets, coupon impedance and loss on the sample defined at engineering, and a cross-section from each lot at the worst-case aspect ratio after three thermal shocks. The measurement methods and their limits are in testing in high-frequency PCB production.

Station 12 — Final Inspection and the Data Package

The last station is not cosmetic. A high-frequency lot leaves with a data package or it leaves with an argument waiting to happen. The package should contain the coupon measurements per layer and per target, the cross-section photographs with the measured barrel thickness and dielectric thickness, the finish thickness by XRF, and a statement of which revision of the stack was built.

The acceptance reference is IPC-A-600G for workmanship and IPC-6015 for performance, plus the coupon limits agreed at engineering. When those three disagree — and they do, because a 20 % barrel void can be a Class 3 reject and a functional pass at the same time — the drawing should say which one governs. A board that meets the customer’s electrical requirement but violates the workmanship standard is still a board that will be questioned in an audit, and the data package is what decides whether the question is answered in a day or in a month.

Cleanliness belongs here too. Ionic contamination and flux residue are not visible and are not part of any impedance measurement, but residue 1–10 µm thick sits directly on the copper surface where the skin depth at 28 GHz is 0.39 µm. On a microwave board, a cleaning step that would be optional on FR-4 is part of the electrical specification. Post-processing, packaging and handling are set out in post-processing in high-frequency PCB production.

The Twelve Windows as One Page

Laid out together, the stations show where the leverage is. Stations 3, 4 and 6 — etch uniformity, lamination pressure and desmear chemistry — account for most of the difference between a build that measures as designed and one that does not, and all three are invisible on the finished surface. Stations 8 and 9 — mask and finish — are the two that can be fixed at the drawing stage for free.

The order also explains a common frustration: a customer who asks for “better process control” is usually asking for tighter tolerances, when the actual gain comes from aligning the windows with the material. A 6:1 aspect ratio limit, a 290 °C press cycle and a sodium naphthalenide desmear are not conservative choices imposed on the design; they are what the laminate allows, and a design that works inside them is a design that yields.

XCE PCB runs this route on conventional and high-frequency builds: 2 to 72 layers, PTFE, hydrocarbon and ceramic laminates, 600×1200 mm panels, impedance held to ±10 %, samples in as fast as 24 hours, inspected to IPC-A-600G and IPC-6015. The decisions that should be settled before station 1 are in considerations in the production process of high-frequency PCB, and the microwave-specific physics that changes these windows is in the production process of microwave PCBs.

Frequently Asked Questions

What tolerance should be put on line width for a high-frequency board?
±10 % or ±25 µm, whichever is tighter, measured at the panel centre and the edge. The two numbers matter for different reasons: ±10 % controls the impedance spread that a designer budgets for, while ±25 µm is the practical capability of the etcher. On a 0.59 mm line the 25 µm figure is 4.2 %, so the percentage governs; on a 0.24 mm line the 25 µm figure is 10 % and both bind at once.
Why can FR-4 and PTFE not be laminated in a single press cycle?
Because the bonding systems cure at different temperatures: FR-4 at 170–190 °C, the RO4000 family at 190–200 °C, and PTFE with FEP film at 290–315 °C. Running the stack at the higher temperature degrades the epoxy; running it at the lower one leaves the FEP unmelted. A mixed stack therefore needs sequential lamination, which re-heats the layers already cured.
What aspect ratio can be plated reliably on PTFE?
About 6:1, against 8:1 to 10:1 on epoxy laminates. The limit comes from three factors together: a non-wetting hole wall, a Z-axis CTE of 237 ppm/°C on 5880 that stretches the barrel by about 82 µm at reflow in a 1.6 mm board, and the thickness of copper that can be thrown into the hole. On a 1.6 mm board, 6:1 puts the smallest reliable finished hole at about 0.27 mm.
Does the lamination pressure really change the impedance?
Yes, on PTFE. The material cold-flows, so finished dielectric thickness depends on the press cycle rather than only on the incoming sheet. A 4 % compression moves a 50 Ω design to 48.8 Ω, and because the compression is uniform across the panel the error appears as a systematic lot-level offset rather than as scatter, which is why it survives incoming inspection.
How is PTFE desmeared?
With sodium naphthalenide or with plasma. Permanganate, the standard epoxy desmear chemistry, does not attack PTFE smear, so a barrel processed that way can look plated and still be disconnected. The result is verified by cross-section after thermal shock, because continuity testing will not reveal a partial void.
Why is the solder mask part of the impedance specification?
Because it is a dielectric layer 20–30 µm thick with a Dk of 3.5–4.5 sitting directly on the trace. Over a 0.254 mm microstrip that lowers the impedance by 1–3 Ω and slightly raises the loss, and the effect is absent from a coupon measured with the mask opened. Defining mask openings over impedance-critical lines and coplanar gaps removes most of the error at no cost in the artwork.
RF & Microwave PCB Fabrication

Need a Build That Holds Every Window?

Send us your Gerber files and requirements for a manufacturability review and a tailored quotation.

2–72 layers ±10% impedance tolerance Manufacturing since 2009